Curvature Preserving Fractal Interpolation Functions: A Hybrid Geometric Approach

Technical Summary: Curvature Preserving Fractal Interpolation Functions

Problem Statement
Fractal Interpolation Functions (FIFs), constructed via Iterated Function Systems (IFS), offer a powerful framework for modeling self-similar and irregular data. However, traditional FIF constructions prioritize self-similarity and recursive flexibility over geometric fidelity, often failing to preserve critical shape characteristics such as curvature. Conversely, classical shape-preserving interpolation techniques (e.g., cubic splines) ensure smoothness and continuity but lack the adaptability to model data with significant irregular, oscillatory, or fractal behavior. Furthermore, classical polynomial splines are unique for a given dataset and often lack the flexibility to capture finer geometric features. There exists a gap in developing interpolation methods that can simultaneously handle irregular data structures while explicitly aligning the interpolant's curvature with that of a reference geometric model.

Methodology
The paper proposes a Curvature-Preserving Cubic Fractal Interpolation Function (CP-CFIF) framework that hybridizes the smoothness of classical cubic splines with the recursive adaptability of fractal perturbations.

1. Construction of the CFIF:
   The method begins with a classical $C^1$-continuous cubic spline $S(x)$ interpolating the given data points $\{(x_t, u_t)\}$. A fractal interpolant $F(x)$ is then constructed using a modified IFS. The vertical mapping functions are defined recursively as:
   $$F(L_s(x)) = \lambda_s F(x) + S(x) + \delta_s(x)$$
   where $L_s(x)$ are affine contractions mapping the domain to subintervals, $\lambda_s \in (-1, 1)$ are local vertical scaling parameters, and $\delta_s(x)$ are correction functions. To ensure $F(x)$ interpolates the data points exactly, $\delta_s(x)$ is defined as $(1-\lambda_s)[S(x) - L^\sharp_s(x)]$, where $L^\sharp_s(x)$ is the linear interpolant between adjacent data points.

2. Penalty-Based Optimization:
   To enforce curvature preservation, the scaling parameters $\lambda_s$ are not chosen arbitrarily but are optimized to minimize the deviation between the curvature of the fractal function ($\kappa_F$) and the reference spline ($\kappa_S$). This is achieved by minimizing a penalty functional:
   $$J(\lambda) = \int_{x_1}^{x_N} (\kappa_F(x) - \kappa_S(x))^2 dx$$
   subject to the constraint $|\lambda_s| < 1$.

3. Theoretical Analysis:
   The paper derives theoretical conditions for the method's validity:

   • Convergence: It is proven that as the scaling parameters $\lambda_s \to 0$, the fractal function $F(x)$ converges uniformly to the classical spline $S(x)$.
   • Curvature Bounds: Theorem 3 establishes sufficient conditions on the magnitude of $\lambda_s$ to ensure the curvature deviation $|\kappa_F - \kappa_S|$ remains within a specified tolerance $\epsilon$. The bound depends on the maximum slope ($K$) and the maximum second derivative ($C$) of the reference spline.
   • Stability: Theorem 4 demonstrates that the curvature of the CP-CFIF is stable under small perturbations of the scaling factors, exhibiting Lipschitz continuity with respect to $\lambda$.

Key Results
Numerical experiments were conducted on three distinct datasets: low-curvature data, high-curvature data (sharp bends), and noisy data (perturbed sine function).

• Accuracy: The proposed method successfully interpolates data points while maintaining curvature profiles closely aligned with the reference cubic spline. In low-curvature scenarios, the deviation is negligible. In high-curvature and noisy scenarios, the method maintains better geometric fidelity than standard splines when scaling parameters are optimized within the theoretical bounds.
• Parameter Sensitivity: Analysis of the scaling factors $\lambda$ (ranging from 0.001 to 0.1) shows that smaller values yield behavior nearly identical to the classical spline, while larger values introduce controlled fractal oscillations. Optimal performance for curvature preservation was observed in the range $\lambda \in [0.01, 0.05]$.
• Computational Cost: The method incurs a computational overhead of approximately 200–250 times that of classical cubic splines due to iterative refinement and curvature optimization. However, absolute execution times remain practical (sub-second for the tested datasets), and the complexity scales approximately linearly ($O(n^{0.16})$) for low-curvature data.

Significance and Claims
The paper claims to introduce a significant advancement in geometric modeling by bridging the gap between fractal interpolation and shape-preserving techniques. The primary significance lies in the ability to generate fractal interpolants that are not only self-similar but also curvature-aware.

By embedding curvature alignment into the interpolation process via a penalty-based optimization, the method enables the effective use of fractal models in applications where shape detail is critical, such as:

• Computer-Aided Design (CAD/CAM)
• Biomedical shape reconstruction
• Image boundary modeling

The authors assert that this approach offers a powerful tool for geometry-aware modeling, balancing mathematical tractability with rigorous curvature preservation guarantees, without requiring the increased parameter complexity of more advanced rational fractal schemes. Future work is identified as extending this framework to rational fractal interpolants, Hermite-type functions, and higher-dimensional surface interpolation.